On angels, demons, and artificial intelligence
I have spent the past few months teaching an undergraduate philosophy seminar on angels and demons, a rather unusual topic for my secular department and university. But I proposed to the class that this subject might light an intellectual path for us to consider another rather more topical area: the nature of artificial intelligence. We therefore spent the first part of the course engaged in a careful study of the Treatise on the Angels in Aquinas’s Summa Theologiae before turning to some classic papers in the philosophy of AI from the 1980s and 1990s, a period of thinking about minds and machines—over which computers went from a hobby to an indispensable part of life for most of us—that continues to shape the intellectual landscape today.[1]
Right from the start of the class, I confessed to being an AI skeptic (at least concerning the technological paradigms for AI that are most prominent now); but I also shared my worry that my views were driven by prejudice, specifically, the prejudice that only human beings or creatures suitably similar to us can have minds. To avoid this prejudice, we must offer a principled account of the basis of intelligence that does not rule out reasonable possibilities for intelligent existence in creatures that are unlike us. Happily, the conversations that the seminar sparked have equipped me with better arguments about why the ever-more-sophisticated chatbots and assistants that we encounter fall short of embodying real intelligence—no matter how good they become at imitating it.
Angels, meanwhile, are commonly depicted as a part of the premodern imagination that ‘science’ or ‘reason or ‘the Enlightenment’ has swept away. What I have come to think is that the angels can help us overcome some deep confusions about the nature of intelligence that the Enlightenment-era mechanical philosophy and its descendants (particularly the computational theory of mind) have encouraged in our own time. Our minds are not simply machines. Nor does a machine count as a mind because it can appear to do some of what intelligent creatures do.
Within the narrower horizon of academic philosophy, we need an adequate theory of intelligence to make this case and avoid the charge of prejudice. But, for our broader cultural conversations, we need imaginative resources that our computational age makes it hard to sustain. As my class found, appreciating the force of the arguments in Aquinas’s Treatise on the Angels required that we free our imaginations in ways that proved fruitful for our philosophical inquiry—but that were also, more generally, liberating.
Scripture is full of stories about the angels. In Genesis and elsewhere in the Torah, angels serve both as visible manifestations of God’s power and as his agents. In later apocalyptic literature, they take on a range of functions, including commanding God’s armies, the angelic host. In the New Testament, angels announce to humankind both the birth of Jesus and his resurrection (as befits their common name, which means ‘messenger’ in both Hebrew and Greek), and thereby spark the story-telling that is central to the Christian mission of spreading the good news. The Christmas story, in particular, is full of angelic presence, and this time of year gives angels their greatest prominence in modern Western culture.
This scriptural-narrative tradition soon blended with a Platonic theological tradition concerning the nature of God and other beings who are above us in the hierarchy of being[2]. For this Platonic tradition, to which Aquinas is an heir, reality is divided between material and intelligible (or ‘spiritual’) things. Human beings are simultaneously the highest creatures in the material order and the lowest creatures in the intelligible order, which consists not of abstract objects of thought but rather of a hierarchy of intelligences. Our intelligence is only a shadow of the intelligence of higher beings such as the angels. Like us, angels are finite, created beings who play a discrete role in God’s providential arrangement of the world, but unlike us they are free of any material limitation.
As I will return to below, it is this conception of angels as immaterial intelligences that makes it useful to us in theorizing the very nature of intelligence: the organic prejudice that appears to beset any account of intelligence that takes human beings and creatures like us as paradigmatically intelligent cannot apply to such a theory. In fact, in this Platonic view, human beings are unique in the intelligible order for being enmeshed in material and sensory life.
But this Platonic philosophical-theological tradition is not the only influence on Aquinas’s Treatise on the Angels. The other main intellectual strand is the Aristotelian theory of mind as a vital power, that is, a power that belongs to and characterizes living things of a specific kind. This theory points in exactly the opposite direction as the Platonic one, since, in the most expansive Aristotelian conception of mindedness, animals too have minds insofar as they have cognitive powers. Moreover, our human mode of mindedness is closely intertwined with our sensory lives.[3]
Aquinas synthesizes these apparently contradictory streams of thought into a single, coherent view of the nature of intelligence. In this view, the core expression of intelligence is the act of understanding, which is the unity of a mind with the very nature of things. We think about individual horses–Bucephalus or Secretariat–but what we understand, if we understand, is equine nature itself. In Aristotle’s hands, this idea offers up the image of the scientist as the paradigmatic knower (an image that is one of his great legacies to Enlightenment and post-Enlightenment ways of thinking that have tended to construe him as their chief premodern antagonist). To speak of understanding as a kind of unification of mind and world, moreover, makes the mind’s work a matter of actively making contact with reality, rather than simply representing things. The mind is not a kind of veil, nor just a means of passively registering or recording the world.
For Aquinas, the angels’ immateriality places them fully in the realm of intelligibility: they understand themselves and the natures of things without mediation and from the moment of their creation onward. For us, understanding is instead a laborious achievement that demands patient inquiry and that must be distinguished from the ordinary knowledge that is provided to us by sources like the senses and testimony. Our mode of intelligence is, in the terms of Aquinas’s theory, discursive, while angelic intelligence is intuitive: all-at-once and complete. In other words, it is not just that the angels are talented scientists (with a remarkable range of specializations: indeed, all of them together!). Rather, they do not need to undertake the work of scientific inquiry at all. There is no barrier at all between them and the natures of things—and so they have innate and internal access to complete understanding.[4]
Still, for the angels, just as it is for us, reflecting on the objects of our understanding is a purposive activity, brought about by the faculty Aquinas calls ‘the will’. In this way, Aquinas preserves Aristotle’s other insight, that cognitive life is proper to animals, since they negotiate their environments, preserve themselves, and pursue their other characteristic goals as purposive self-movers of a special sort. To be sure, intelligence is a higher and more perfect kind of self-movement than mere animal self-movement, one that allows intelligent creatures like us to cognize their purposes generally and to order their various pursuits on the basis of what they conceive to be good for them overall. But there are important continuities that undermine any sharp ontological divide between material and spiritual reality. Aquinas’s world is not Descartes’s, divided absolutely into active mind and inert matter, a division that would (among other baleful consequences) destroy the unitary nature of human beings as intelligent animals.
At first glance, large language models (LLMs) and other forms of so-called generative AI based on the ‘transformer’ architecture seem to resemble angelic intelligence. A lovely essay by Joe Moran, a professor of English and cultural history, appeared earlier this year entitled “You are not an angel (a letter to my students)”. Moran charts the history of associating the instantaneity of technological communication with angelic thought before making the case that the time-intensive discursive thinking that reading and writing embody must be distinguished from the generation of content that LLMs do so successfully. Moran argues that since we “will never be angels [...] [w]e must roll up our sleeves and get to work – the work of making meaning together with language, one word at a time.”
The epistolary form that Moran employs and his choice of recipient–his students, many of whom are now habitually using this technology as a crutch or a substitute for their own intellectual work–is particularly apt. After all, ChatGPT cannot really write a letter. To be sure, it can compose a text that resembles a letter, much more quickly than any of us can. But to write a letter is to take a certain kind of journey, to choose–and continue to choose–one’s words reflectively, to subtly alter one’s course in the middle of the way as one sees more and better the terrain that lies ahead, and to choose an apt stopping point that is typically just the middle of a longer correspondence (or simply the point that one runs out of ink or writing paper). It may take a brief moment for an LLM-powered chatbot to summon the computational resources to deliver a complete answer, but there is nothing truly discursive in what it does.
In a striking passage in the Treatise on the Angels, Aquinas argues that the angels do not really speak to human beings. My students found this claim to be rather surprising, since they knew that Aquinas treats scripture with great care and scripture seems to be full of examples of angels speaking in their capacity as messengers. Think of the Annunciation of Gabriel to the Virgin Mary when he reassures her, “Do not be afraid” (Luke 1:30). But in Aquinas’s view, speech is a vital activity, one that is not only discursive but also embodied. According to Aquinas, what really happens during such angelic visitations is that the angels simulate speech-sounds so that human beings can grasp something that is, taken in itself, purely intelligible. Angels, then, engage in just the sort of simulacrum of discursive thinking that LLM output appears to exemplify.
Do these analogies suggest that algorithmic intelligence is (or someday soon will be) superior to human intelligence, just as angelic understanding outstrips human discursive thought? Most of the temptation to think so comes from the computational model of mind, a thought-picture that describes intelligence in terms of algorithmic operations on a given input that lead to corresponding output. Whether this output is linguistic or behavioral (an action or other change-of-state abstractly describable in a space of possibilities), it may seem that human beings, at least some non-human animals, and computational implementations of algorithms equally meet the standard for possessing intelligence.
The hybrid Platonic-Aristotelian theory that Aquinas develops presents us with an instructive alternative: mindedness as freedom from material limitation. There are two dimensions to this freedom, which correspond to the two core faculties of mind: intellect and will.
The first dimension of freedom belongs to the intellect, the capacity to understand not simply the particularities of the world that confront us–which we encounter in the stream of our own consciousness–but to grasp any intelligible reality whatsoever. In this way, we stand apart from the individual encounters with the world that our senses–and even our imaginative capacities–afford. At the same time, the part of the world we access through our minds is really there, not just as a realm of mental concepts, but as the structuring nature of reality. The existence of such natures in our minds when they are understood is, for Aquinas, a sign of our freedom from material constraint. Think, by contrast, of the image of the mind as a wax tablet receiving impressions from what we encounter[5]. If our minds were simply stamped by the world, we would have access only to the surfaces of things, the aspects communicated via the senses.
The second dimension of freedom belongs to the will, by which we direct our intellectual lives–and, indeed, all the dimensions of our lives that we can consciously shape. There is no such thing as a mind without a will, just as there cannot be an animal that is purely passive in its environment. Mindedness is a form of responsiveness, in other words. Moreover, we set our own goals; they are not simply given to us. Intelligent life cannot be instinctual, even if the way that a thing responds to stimuli is highly sophisticated. Many things have purposes that are not purposive in this way, artifacts and social institutions among them.
To be clear, Aquinas thinks of human intelligence, at least, as depending essentially on our embodiment, even as the operations of intellect and will are free of material limitation. We do not think with some spooky part of ourselves, an angel within. Rather, his point is that our capacities for self-directed intelligent activity manifest a kind of independence from the physical pushes and pulls of the world, an independence that the other animals share in a lesser way.
LLM-powered artificial intelligence has neither kind of freedom. On the side of the intellect, such so-called ‘AI’ is capable neither of discursive thought nor of understanding, since it does not aim to grasp the world at all. Rather, it merely synthesizes our representations. Just as an echo in a cave is not the speech of rocks, the deliverances of LLMs are only simulacra of language–and the thought that ultimately lies behind it. On the side of the will, an LLM simply has no purposes of its own, let alone ones that it could cognize and order.
The lesson here is that it is easy, in some circumstances, to confuse the intelligence needed to use a tool with the intelligence of the tool itself. We tend not to make this mistake with saucepans and tables, but the greater the hiddenness of a thing’s mechanism, the more it is tempting to project purposiveness. The ‘black box’ quality of LLMs makes them even an especially tempting site of such projection.
I noted above that angels and demons are commonly associated with a premodern world-view that reason or science are meant to have swept away. In fact, it is the technological world-view of AI boosterism that is prone to a form of irrationalism: attributing to pieces of technology or algorithms apocalyptic power.
One reason to adopt a clear-eyed view of so-called AI is to hold its creators account for the social challenges that the technology has already caused and is likely to cause in the future. The problems do not lie with some hypothesized difficulty of AI ‘alignment’ but rather with the all-too-human situation of a largely deregulated industry run by people who take childlike glee in ‘disruption’. Not only do I remain an AI skeptic in the sense of thinking that the technology in its present guise offers only a simulacrum of intelligence, I am also skeptical that the industry is on a path to taking genuine responsibility for its effects. It is up to us to push back.
I have argued that Aquinas’s theory of intelligence offers us both argumentative and imaginative resources to respond to these intellectual mistakes. That is not the only reason to study Aquinas, of course, but I am increasingly persuaded that we must foster a culture of humanism, which seeks wisdom with intellectual humility in the company of the vast set of texts and traditions handed down to us, in the face of a rising technological philistinism that pretends to have all the answers. Nor should we humanists close ourselves off to the pressing issues of the day or cede what is properly our own intellectual territory.
[1] The Treatise on the Angels is contained in the Prima Pars, Questions 50–64. We used Alfred Freddoso’s wonderful translation of the Summa available on his website, which renders the text in a vivid and unfussy register. Saint Thomas Aquinas, of course, is known as the ‘Angelic Doctor’ in Roman Catholic theology, not least for his contributions to the study of this topic.
[2] On this subject, the 5th century text On the Celestial Hierarchy (traditionally attributed to Dionysius the Areopagite) is the Christian-Platonist source par excellence. [Pseudo-]Dionysius describes in detail the orders and ranks of angels but insists on absolute immateriality as a quality all of them share.
[3] I have written before for The Line of Beauty about what I find most attractive in this theory of mind and what the analogies—and continuities—between animal and human intelligence can help us see about ourselves.
[4] Strictly speaking, the things that the angels understand (the ‘intelligible species’ they have from the moment of their creation) are derived from the Divine Ideas, the types of all things that exist in God’s mind and from which the reality of things proceeds when he gives them their being. The angels are glistering mirrors of divine knowledge.
[5] The image of the wax tablet is ultimately drawn from Plato’s Theaetetus, but enjoyed a long afterlife via Stoic philosophy.
What an interesting and revealing analysis. It helps elucidate the philosophical issues clouding the proper use of this new technology.
Thanks for this fantastic essay breaking down, among other things, why AI is not truly “intelligent” and does not—and cannot—employ embodied discursive thinking or “angelic” intuition. I really enjoyed this piece.